Module talk:Footnotes

Jump to navigation Jump to search

False positives with Template:Jewish Encyclopedia[edit]

As seen here: Baal#cite_note-FOOTNOTEKohler1902-87 – Finnusertop (talkcontribs) 10:19, 1 March 2021 (UTC)

 Done
Trappist the monk (talk) 17:16, 22 April 2021 (UTC)

False positives with Template:New Cambridge Medieval History[edit]

As sen here: Byzantine Empire#cite_note-Louth_2005_113–115-54 – Finnusertop (talkcontribs) 06:14, 3 March 2021 (UTC)

 Done
Trappist the monk (talk) 17:17, 22 April 2021 (UTC)

False positive with "New Cambridge Medieval History"[edit]

{{New Cambridge Medieval History}} – see e.g. Ayyubid dynasty#References. Is anybody actually monitoring this page? Wham2001 (talk) 19:45, 17 March 2021 (UTC)

 Done
Trappist the monk (talk) 17:17, 22 April 2021 (UTC)
@Trappist the monk: basically all refs here are false positives. – Finnusertop (talkcontribs) 14:47, 25 April 2021 (UTC)
@Wham2001 and Finnusertop: {{New Cambridge Medieval History}} is one of the templates where, in the absence of a specific |last=, the "lastname" part of the citeref comes from an |editor-last= buried in the template encoding, same for year: in this case (and many similar) it's indexed by the volume number. So rather than adding every editor/year as a separate entry in the whitelist itself, a more accurate approach would be to recognize the volume-dependent default using an appropriately indexed table in the module's data. Mark you, this is a vague memory from a year ago, and I thought Trappist the monk had implemented the construct, at least in part, so I may be missing something. And, of course, easy for me to say especially as I don't speak LUA. David Brooks (talk) 19:02, 25 April 2021 (UTC)

False positives with "cite RNB1823" and "Cite NBD1849"[edit]

Please add {{cite RNB1823}} (Marshall) and {{Cite NBD1849}} (O'Byrne) to this list. — PBS (talk) 15:55, 22 April 2021 (UTC)

Both can have the year or the date overide the default year because there may be multiple instances of references to the same year necessitating YYYYa-z formats.

{{Cite NBD1849}} (O'Byrne) defaults to 1849

{{cite RNB1823}} (Marshall) uses the following script for its defaults.

                 {{#switch:{{{volume}}}
                 | 1 = 1823
                 | 2 = {{#switch:{{{part}}}|1 = 1824 |2 = 1825 |1824–1825}}
                 | 3 = {{#switch:{{{part}}}|1 = 1831 |2 = 1832 |1831–1832}}
                 | 4 = {{#switch:{{{part}}}|1 = 1833 |2 = 1835 |1833–1835}}
                 | sup |sup. |supp |supp. |supplement |supplemental = {{#switch:{{{part}}}|1 = 1827 |2 = 1828 |3 = 1829 | 4 =1830 |1827–1830}}
                 | <!-- default if no volume specified -->1823–1835

PBS (talk) 16:34, 22 April 2021 (UTC)

Two more wrappers "Cite Colledge2006" and "Cite Colledge2006"[edit]

Two more templates for the list {{Cite Colledge2006}} and {{Cite Colledge2010}}PBS (talk) 16:47, 22 April 2021 (UTC)

Other templates[edit]

see previous discussion: Module talk:Footnotes/Archive 1#Dynamic whitelist revisited

@user:Trappist the monk there seems to be a backlog here. I have a score of other wrapper templates I could list here, there has to be a better way to do this. — PBS (talk) 16:02, 22 April 2021 (UTC)

For more see Wikipedia:WikiProject Wikisource/Citation Uniformity and User:PBS/Notes#List of PD TemplatesPBS (talk) 16:07, 22 April 2021 (UTC)

If anyone has found a better way to do this, they have not chosen to voice their findings here. Do you have a better way?
Trappist the monk (talk) 16:52, 22 April 2021 (UTC)
I only became involved because of a discussion on a user's talk page (User talk:Keith D#Harv error - false positive), until that moment I had no idea that such tracking took place and I still do not understand its purpose. It just seems to me, with my superficial knowledge, to be a Sisyphus task for someone to maintain with all the thanks that goes with such a commitment.
Perhapse you can point me to a page that explains why perfectly good long citation that is linked to with a short citations using a Harvard template need to be listed here because of false positives. -- PBS (talk) 17:09, 22 April 2021 (UTC)
To be fair, I did start a discussion on a possible more flexible solution about a year ago (Dynamic whitelist revisted) but I do recognize that (a) the universe of templates contains a lot of complexities that can defeat a programmatic approach (b) I'm basically saying "easy for you to do" when I realize you're a frantically busy guy. I've pretty much forgotten the arguments I put forward then because you scratched my particular itch (thank you!), but I remember you did add valuable input. David Brooks (talk) 17:18, 22 April 2021 (UTC)
Category:Harv_and_Sfn_template_errors is the only thing that I have written about this issue. The whole mechanism was conceived as a way to inform editors that an article has broken short-ref / long-ref links. Most editors do not use one of the scripts that highlight these errors so the vast majority of us are wholly unaware of these broken linkages unless we take it upon ourselves to test each link individually. Most of us don't even think about that because, after all, those links are blue, right? When we first turned these error messages on, the community, as it is wont to do, overreacted because, you know, we hate, hate, hate change, so the error messaging is hidden. As a result of no one knowing that the links are broken, there are still 26k articles in Category:Harv and Sfn no-target errors‎ (down only a little from 29k in May 2020) and 3100 articles in Category:Harv and Sfn multiple-target errors‎ (down a tad more from 4600 at the same sample point); see this archive.org snapshot.
Trappist the monk (talk) 17:52, 22 April 2021 (UTC)
Thank you User:DavidBrooks that link to an old conversation was a great help. -- PBS (talk) 11:30, 23 April 2021 (UTC)
I still think that this subsystem is an inelegant solution because it requires continuous manual supervision and amendment, although I can not see an obvious alternative given that some think the ends justify the means.
One of the problems with the current system is edits like this. It will "fix" the particular page, but it is not clear unless you are in the know why it is necessary, and every new instance will need it. However that fix has three problems. The first is that other editors will remove the extra parameter as it is not obvious what it is there for (as it is hiding a false positive entry in a maintenance category). The second is more of a problem, as once in place, the extra parameter ignore-err=yes can create false negatives if in the future any information is added to the long citation (such as a change in date for a different edition). The third is that it is likely that new instances will be created over time and that editors will have to repeatedly make similar edits to new instances in other articles. I think the solution to this is to improve the documentation for editors to report two or more false positives here if they involve non-standard citation templates rather than plastering over the cracks. -- PBS (talk) 11:30, 23 April 2021 (UTC)
If you know how the documentation can be improved, please do so.
Trappist the monk (talk) 16:47, 23 April 2021 (UTC)
I remember thanks to recent additions to the whitelist data infrastructure that Trappist the monk did indeed implement support for that kind of scheme - thanks. David Brooks (talk) 21:17, 27 April 2021 (UTC)
Trappist the monk, thank-you very much for the additions to the whitelist. I have been working through Category:Harv and Sfn multiple-target errors‎ for a few months now, systematically fixing the more straightforward cases. At the current rate it will take me about another year to get through to the end and the category should be left with 500 articles or so. I think that there are a few other editors also working on the same backlog. The no-target errors, though, are a more daunting backlog. I think it would be very good if it were possible to have the error messages display by default, and I would hope that the community might be a bit more receptive if they all represent "real" errors. Best, Wham2001 (talk) 21:09, 26 April 2021 (UTC)
I think that we could turn on the multiple-target errors without too many pitchforks launched our way but I'm not willing to turn on the no-target error messages and I certainly won't be the one to cleanse Category:Harv and Sfn no-target errors‎ of all of the false positives so those errors are going to remain hidden. Unless... I have often wondered what would happen if we metered error messaging – at first only emit error messages in articles with names ending in a certain letter; some time period later, add articles that end with another letter, etc until finally all are shown – sort of the frog-in-the-cookpot version of error messaging. So, should we turn on the multiple-target errors and see what happens?
Trappist the monk (talk) 23:44, 26 April 2021 (UTC)
I would support turning on the multiple-target errors, though it's likely you that the people with pitchforks will come after! Another couple of thoughts: (a) is it possible to identify templates transcluded from pages that also transclude {{sfn}} / {{harv}} / etc. and which themselves transclude a CS1/2 template? That would, as I understand it, provide a pile of transclusions to look through for false positive errors. (b) Would it be helpful to advertise your script (which as I understand it emits very few false positives) on the documentation pages for {{sfn}} etc? I suspect that a big part of the problem is editors thinking that if the link in the short footnote is blue then it must be working (by analogy to wikilinks). Best, Wham2001 (talk) 07:58, 27 April 2021 (UTC)
Multiple-target error messages now showing. At this moment, Category:Harv and Sfn multiple-target errors has 3066 members. I'm afraid that your question (a) is sufficiently convoluted that I don't think that I can understand it well enough to give a rational answer. All of the ways to show errors are named at Category:Harv and Sfn template errors which is linked from the error message help link. I have thought the same which is why I wrote above Most of us don't even think about that because, after all, those links are blue, right?
Trappist the monk (talk) 13:03, 27 April 2021 (UTC)
User:DavidBrooks/EncycStats might be a sort of answer to Wham2001's question. It hasn't been updated in ten months, but I suspect that it is still a useful list. The "articles" links will take you to pages that transclude a given template and have at least one error (not necessarily from that template). I cleaned out most of User:DavidBrooks/CiteBookStats last year, since templates transcluding {{cite book}} typically have only one author. Templates transcluding {{cite encyclopedia}} are more challenging, however, since the list of authors can be endless. – Jonesey95 (talk) 14:13, 27 April 2021 (UTC)
Yes, I think that list probably captures all the templates I was thinking of. Thanks! I shall bookmark it and come back to it once I've worked through the multiple-target errors. Wham2001 (talk) 19:22, 27 April 2021 (UTC)
I'm flattered those lists are still useful. It did require a fair amount of API spelunking. But I've also noted that {{Cite web}} has been used by editors in the past to achieve similar (erroneous) ends—after all, {{Cite web}} can be used to reference any online encyclopedia. David Brooks (talk) 21:17, 27 April 2021 (UTC)
That (or some other action?) seems to have had the side-effect of adding user sandboxes to the category – at least, a few have suddenly appeared in the first 200 pages... I guess they can probably be left alone for the moment, though. Wham2001 (talk) 19:27, 27 April 2021 (UTC)
Trappist the monk, after a week and a half of error messages being enabled, I've seen numerous editors who aren't primarily reference gnomes fixing these errors, and new articles are being introduced into the category at a lower rate (~1/day rather than a few/day beforehand). I consider that a win.
On the other hand, I have found what I think is a false positive, or, at least, I cannot work out where the problem is. History of LGBT characters in animation: 2010s shows a multiple-target error for {{sfn|GLAAD|2017}}. There is only one source emitting that anchor id (I can only find it once in the rendered page's html) and if I comment the source out the error changes to a "no-target" error. The error appeared with this diff. Any ideas? Thanks, Wham2001 (talk) 11:56, 9 May 2021 (UTC)
Fixed. I found the problem by doing a Find for "GLAAD" on the page and looking for sources dated 2017. – Jonesey95 (talk) 14:02, 9 May 2021 (UTC)
(edit conflict)
This one. {{cite tweet}} invokes Module:Cite tweet which assembles |author= from one or more of |last=, |last1=, |first=, |first1=, |author=, |author1=, |author-link=, |user= plus some static text. The article reader is not sufficiently sophisticated to know what metatemplates do to assemble their anchor IDs in all of the ways that they do so the article reader only looks for |last= aliases and |year= or |date=. In this case, it sees |author=GLAAD and |date=May 25, 2017 and creates CITEREFGLAAD2017 which is also created by this {{cite report}} which uses |ref={{sfnref|GLAAD|2017}}.
At Template:Cite tweet § Using {{sfn}} and {{harvid}} users are instructed to use {{harvid}} in |ref= when using {{sfn}}. It seems to me that Module:Cite tweet should be tweaked to set |ref=none ...
Trappist the monk (talk) 14:11, 9 May 2021 (UTC)
Ohhhh, I see. Thank-you both very much! I agree that it would be more consistent with the documentation for {{cite tweet}} to not produce an anchor by default (I'm also not convinced that we should be encouraging editors to cite tweets in the first place, but that's a different argument!). Thanks, Wham2001 (talk) 15:28, 9 May 2021 (UTC)
When we first turned these error messages on, the community, as it is wont to do, overreacted because, you know, we hate, hate, hate change, so the error messaging is hidden.
The community didn't overeact, the way "error" detection is inherently flawed and had a false positive rates in the range of ~10%, sometimes turning every page on a single topic littered with red warnings about something catastrophic when everything was peachy. And that is simply not acceptable. Additionally a plethora of errors would have been fixed by simply making |ref=harv default (which it now is), as well as thousands of errors being bot-fixable. If that legwork at been done before, and if the errors flagged were all actually errors, it would have been a different story. But that wasn't the case, and still isn't.
See Module_talk:Footnotes/Archive_1#Error_messages and the few discussions above and below for the full context. Headbomb {t · c · p · b} 13:02, 3 May 2021 (UTC)

Whitelist problem[edit]

An edit on 23:31, 25 April 2021 at Module:Footnotes/whitelist causes Lua error in Module:Footnotes/anchor_id_list at line 503: attempt to index field '?' (a nil value). (four times) for ref 2 at Reconquista. Ref 2 is:

{{harvnb|Collins|1989|p=147}}
{{harvnb|Reilly|1993|pp=75–76}}
{{harvnb|Deyermond|1985|p=346}}
{{harvnb|Hillgarth|2009|p=66 n. 28}}

I'm sure this is a simple problem but it needs someone familiar with the system. Johnuniq (talk) 03:25, 26 April 2021 (UTC)

Fixed, thanks.
Trappist the monk (talk) 12:57, 26 April 2021 (UTC)

Emit category/error messages when 1= numbers only[edit]

For example, [1] should produce a visible error, I feel. Headbomb {t · c · p · b} 22:00, 9 May 2021 (UTC)

Likewise if |1= is empty, e.g. [2]. Headbomb {t · c · p · b} 22:03, 9 May 2021 (UTC)

References

  1. ^ 1933, p. 6-7.
  2. ^ & Smith (2000).
The second one definitely looks like an error to me. As for the first, I can imagine a legitimate case in which someone wanted to cite a no-byline article from 2600: The Hacker Quarterly without a date for some reason, creating CITEREF2600. A database dump report should be able to find instances of sfn and sfn-like templates that use only numbers in the first parameter, and which pages they appear on. That would tell us whether there are edge cases that we are not thinking of. – Jonesey95 (talk) 22:13, 9 May 2021 (UTC)
Follow-up: I found an actual legitimate example of all-numeric |1= in the wild, in Tektronix 4010. The vast majority of them are errors, though, so we might want to flag it as an error and explain that in the rare edge case, editors could work around it using a literal |1=. – Jonesey95 (talk) 22:32, 9 May 2021 (UTC)
Fixed the Tektronix 4010 ones to produce something legible. Headbomb {t · c · p · b} 22:44, 9 May 2021 (UTC)
Looks like a good fix. I found one other article like that in the search linked above. I support marking these as errors, since there are multiple good workarounds and 95+% of the instances are errors. Here's a link to the same search with the harv template. – Jonesey95 (talk) 00:48, 10 May 2021 (UTC)